CN112914527B - Arterial blood pressure signal acquisition method based on pulse wave photoplethysmography - Google Patents
Arterial blood pressure signal acquisition method based on pulse wave photoplethysmography Download PDFInfo
- Publication number
- CN112914527B CN112914527B CN202110184194.5A CN202110184194A CN112914527B CN 112914527 B CN112914527 B CN 112914527B CN 202110184194 A CN202110184194 A CN 202110184194A CN 112914527 B CN112914527 B CN 112914527B
- Authority
- CN
- China
- Prior art keywords
- layer
- downsampling
- blood pressure
- level
- arterial blood
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
- A61B5/021—Measuring pressure in heart or blood vessels
- A61B5/02108—Measuring pressure in heart or blood vessels from analysis of pulse wave characteristics
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Abstract
The invention discloses an arterial blood pressure signal acquisition method based on pulse wave photoplethysmography, which comprises the following steps: firstly, constructing a mapping from pulse waves to arterial blood pressure waveforms, specifically adopting a neural network based on Wave-U-Net as a basic framework, wherein the whole network comprises: a downsampling path, an intermediate layer, an upsampling path, and an output layer; training the network, adjusting parameters to obtain an optimal model for converting the pulse wave signals into arterial blood pressure signals, wherein the training comprises the following steps: data acquisition, data preprocessing and network training. According to the invention, the noninvasive wearable pulse wave signals obtained from the fingertip positions are converted into high-quality arterial blood pressure signals which can be obtained by invasive invasion at present through the deep neural network, so that the complexity of collecting the high-quality arterial blood pressure signals is reduced, the possibility of noninvasive monitoring of the arterial blood pressure signals for a long time is provided, and the development of mobile health monitoring technology is promoted.
Description
Technical Field
The invention relates to the technical field of noninvasive sustainable physiological signal analysis and acquisition, in particular to an arterial blood pressure signal acquisition method based on pulse wave photoplethysmography.
Background
In recent years, with the development of sensor technology, a blood pressure monitoring method based on photoplethysmography (PPG) has become a big research hotspot. The PPG technology utilizes an LED light source and a photoelectric sensor to measure attenuated light reflected and absorbed by human blood vessels and tissues, trace the pulsating state of the blood vessels and measure pulse waves. Such pulse waves are related to the periodic activity of the heart and have extremely high correlation with arterial blood pressure signals. In the field of mobile health monitoring, researchers have developed a number of sustainable blood pressure monitoring methods using PPG technology.
One problem with current methods of using pulse wave signals to gather blood pressure information is that: high quality continuous arterial blood pressure waveform signals cannot be obtained with pulse waves. The existing methods for acquiring blood pressure signals based on pulse waves are generally divided into a method based on feature extraction and an end-to-end deep learning method.
The feature extraction-based method requires researchers to perform feature extraction on pulse wave signals, first derivative velocity plethysmograph waves (velocity plethysmography, VPG) of pulse waves and second derivative acceleration plethysmograph waves (acceleration plethysmography, APG) of PPG, and then send a plurality of extracted features into a regression model to predict blood pressure levels. The drawbacks of this type of method are quite apparent. Firstly, the characteristic extraction process inevitably causes information loss; secondly, the method has higher requirements on the quality of signals, and for some signals with lower quality, many features have larger errors and even cannot be extracted; finally, this type of method can only obtain a mapping process from pulse wave characteristics to certain fixed blood pressure characteristics, and the obtained information is very limited.
The end-to-end deep learning method is improved on the basis of a feature extraction method, and utilizes the rapid and powerful processing capacity of deep learning to complete the complex feature extraction process by neural network learning, so that the mapping from pulse wave signals to physiological parameters is directly realized. The method has the advantages of low calculation cost, low requirement on signal quality and the like. However, the existing end-to-end method does not overcome the disadvantage that the feature extraction-based method can only obtain several fixed physiological parameters, and cannot extract continuous arterial blood pressure waveforms from pulse wave signals. In addition, due to the characteristics of the deep learning method, once the neural network is generated, the number of the output characteristics of the neural network is difficult to change, so that a large amount of information in the pulse wave signal can be mapped to a limited number of characteristics, and once the user demand is changed, the original model is abandoned, and the popularization of the PPG technology in the field of mobile health monitoring is limited.
Disclosure of Invention
The invention aims to avoid the defects of the prior art, and provides an arterial blood pressure signal acquisition method based on pulse wave photoplethysmography, so that the accurate arterial blood pressure signal can be acquired by directly utilizing the pulse wave signal, the complexity of the high-quality arterial blood pressure signal acquisition process is reduced, and the accuracy and the robustness of blood pressure information acquisition in the field of mobile health monitoring are improved.
The invention adopts the following scheme for solving the technical problems:
the invention relates to an arterial blood pressure signal acquisition method based on pulse wave photoplethysmography, which is characterized by comprising the following steps:
step one, constructing a neural network taking Wave-U-Net as an infrastructure, which comprises the following steps: a downsampling path, an intermediate layer, an upsampling path, and an output layer;
step 1.1, the downsampling path is formed by cascading M downsampling modules, and the M downsampling modules are respectively marked as Down Block 1 ,...,DownBlock m ,...,DownBlock M The method comprises the steps of carrying out a first treatment on the surface of the Wherein, down Block m Represents an M-th level downsampling module, m=1, 2,; down block of m-th level downsampling module m DSKConv1d as m-th level one-dimensional downsampling adaptive convolution kernel convolution layer m And Downsampling layer Downsampling at level m m Composition;
m-th level one-dimensional self-adaptive convolution kernel convolution layer DSKConv1d m Comprising the following steps: an mth separate convolution layer, an mth compressed re-excitation processing layer and an mth weighted combination layer, and an mth one-dimensional downsampling adaptive convolution kernel convolution layer DSKConv1d m The output of (a) is denoted as DSKreult m ；
The mth separate convolution layer includes: x number of parallelSeparate convolutions of rows, X batch normalization layers, and X layers of LeakyReLU activation functions with slope α; wherein the size of the convolution kernel of the xth separate convolution is F x And a convolution kernel F which is separately convolved with the (x+1) th one x+1 The difference between is k, x=0, 1,..x-1; let initialize F 0 ＝p；
Downsampling layer Downsampling of m-th level m For changing the time resolution of the signal to be original
r represents a scaling factor; downsampling layer Downsampling of m-th level m The output of (2) is denoted as Down result m ；
Step 1.2, the up-sampling path is formed by cascading M up-sampling modules, and the M up-sampling modules are respectively marked as UpBlock 1 ,...,UpBlock m ,...,UpBlock M The method comprises the steps of carrying out a first treatment on the surface of the Wherein UpBlock m Representing an mth level upsampling module; m-th level up-sampling module UpBlock m From Upsampling at the mth level m Sampling layer and m-th level one-dimensional up-sampling self-adaptive convolution kernel convolution layer USKConv1d m Composition; the m-th level one-dimensional up-sampling self-adaptive convolution kernel convolution layer USKConv1d m And m-th level one-dimensional downsampling adaptive convolution kernel convolution layer DSKConv1d in downsampling path m Similarly, m-th level one-dimensional upsampling adaptive convolution kernel convolution layer USKConv1d m The output of (2) is denoted as USKreult m ；
M-th level Upsampling layer Upsampling m The time resolution of the signal is changed into the original r times by adopting a linear interpolation method, and the up-sampling layer of the mth level is Upsampling m The output of (2) is denoted as Upresult m ；
Step 1.3, the intermediate layer is a one-dimensional convolution layer, comprising: a one-dimensional convolution with a convolution kernel size q, a batch normalization layer, and a LeakyReLU activation function layer with a slope alpha; the output of the intermediate layer is denoted as Mresult;
the output layer is another one-dimensional convolution layer, comprising: a one-dimensional convolution with a convolution kernel of size q, a batch normalization layer, and a tanh activation function layer;the output of the output layer is marked as ABP est Namely arterial blood pressure signals;
step 1.4, inputting an input signal into the neural network, sequentially processing a downsampling path, an intermediate layer, an upsampling path and an output layer, and outputting an arterial blood pressure signal ABP est The method comprises the steps of carrying out a first treatment on the surface of the Wherein:
when m=1, 2,..m-1, the M-th stage downsampling module in the downsampling path downsamples down block m In (1) an m-th level one-dimensional downsampling adaptive convolution kernel convolution layer DSKConv1d m Output of DSKreult m Except Downsampling as an mth level Downsampling layer m In addition to the input of (2), the up-sampling module UpBlock of the (m+1) th stage m+1 In (1) m+1st stage one-dimensional upsampling adaptive convolution kernel convolution layer USKConv1d m Output USKreult of (C) m+1 Together as an m-th up-sampling module UpBlock m Middle m-th level one-dimensional self-adaptive convolution kernel convolution layer USKConv1d m Is input to the computer;
m-th downsampling module Down Block in downsampling path M Middle M-th level one-dimensional downsampling self-adaptive convolution kernel convolution layer DSKConv1d M Output of DSKreult M Except Downsampling as an M-th Downsampling layer M In addition to the input of (2), the M-stage up-sampling module UpBlock is used together with the output Mresult of the middle layer M Mid M-th level one-dimensional up-sampling self-adaptive convolution kernel convolution layer USKConv1d M Is input to the computer;
step 1.5, establishing a loss function L by using the formula (1) and taking the loss function L as an optimization target of the convolutional neural network:
L＝L mse +λL max (1)
in the formula (1), L mse Is the mean square loss and is obtained by the formula (2); l (L) max Is a maximum loss function and is obtained by the formula (3); lambda is L max Weights of (2);
in the formula (2), L is the number of signal sample points;
arterial blood pressure waveform ABP representing network generation est Is the first sample point of (2); />
A first sample point representing a reference arterial blood pressure waveform;
in the formula (3), max is a maximum value taking function;
training a network to obtain an optimal model for converting pulse wave signals into arterial blood pressure signals;
step 2.1, acquiring a fingertip pulse wave signal P with a processing window length L and an arterial blood pressure signal A acquired by an invasive blood pressure sensor from a database;
step 2.2, preprocessing data;
step 2.2.1, performing trend removal and normalization treatment on the pulse wave signal P to obtain a preprocessed pulse wave signal
Then +.>
Obtaining a first derivative velocity plethysmograph signal by taking the first and second order differences>
And second derivative acceleration plethysmograph wave +.>
Finally, the preprocessed pulse wave signal is +.>
First derivative velocity plethysmographic signal +.>
And second derivative acceleration plethysmograph wave +.>
Merging the signals into a 3 xL matrix and taking the matrix as an input signal IN of a neural network;
step 2.2.2, normalizing the arterial blood pressure signal A to obtain a normalized arterial blood pressure signal ABP ref And normalized arterial blood pressure signal ABP ref A synchronization reference signal of length L as an input signal IN;
step 2.3, training the network;
input signal IN is sent into neural network, and corresponding reference signal ABP ref Calculating a loss function L as an optimization target, enabling the loss function L to tend to be stable in the training process, and stopping training and obtaining an optimal model when the absolute value of the variation delta L of the loss function L is always smaller than a set threshold value theta;
step 2.4, arterial blood pressure signal post-processing:
arterial blood pressure signal ABP output by optimal model est And the amplitude of the arterial blood pressure signal is restored to the original dimension, so that the arterial blood pressure signal after conversion is obtained.
Compared with the prior art, the invention has the beneficial effects that:
1. according to the invention, the pulse Wave signals are directly mapped to the arterial blood pressure signals through the neural network based on the Wave-U-Net, so that the process of manual feature extraction is avoided, and the information loss caused by the manual feature extraction is avoided. Wave-U-Net is a neural network structure specially aiming at one-dimensional signal processing, and has achieved a certain result in the fields of voice signal separation and denoising. Compared with a traditional time sequence processing network Long-Short time memory network (LSTM) (Long Short-terrm Memory Networks) and a convolutional neural network (Convolutional Neural Networks, CNN), the downsampling layer of the Wave-U-Net can acutely capture the characteristics of different scales in one-dimensional signals, so that the information of input signals is more fully utilized, and more accurate results are obtained. The conversion from PPG signal to ABP signal is essentially the mapping relation of two one-dimensional signals with higher correlation, and the mapping relation can be well obtained by using the Wave-U-Net architecture.
2. The invention fully considers the influence of the input signal on the learning process of the neural network, carries out special design on the input of the neural network, and provides richer information for the network on the premise of not changing the network structure. The pulse wave signal, the velocity plethysmograph wave signal and the acceleration plethysmograph wave signal are simultaneously used as the input of the neural network, so that the time domain information contained in the PPG signal is fully utilized, the difficulty of network learning mapping from the PPG signal to the ABP signal is reduced, and the accuracy of outputting the ABP waveform is further improved.
3. Aiming at the characteristics of waveform mapping, the invention adds a new loss function for improving the quality of the local characteristics of the waveform on the basis of the traditional loss function, so that the ABP waveform generated by the network is highly consistent with the invasive ABP waveform in the overall shape and phase, can keep close to the invasive ABP waveform in specific details, and better retains the shape information of the ABP signal. In addition to the conventional mean square error loss, the present invention adds a maximum loss to the loss function of the network. The optimization objective of the maximum loss is to reduce the maximum of the error between the output waveform and the reference waveform. In the case of using the mean square error loss alone, the output waveform of the network can better approach the reference waveform at a generally trend and flatter position, but in the locally sharp portion of the waveform, the mean square error loss alone cannot obtain a good effect. The maximum loss has a great punishment effect on the error of the sharp part of the waveform, and can well improve the output quality of the output waveform at the sharp part, thereby improving the accuracy of the output ABP waveform.
4. In the up-sampling path and the down-sampling path, a one-dimensional self-adaptive convolution kernel convolution layer is used for replacing a one-dimensional convolution layer in a traditional Wave-U-Net structure. The convolution kernel of the conventional one-dimensional convolution layer is fixed, which results in that the receptive field of the whole network is limited to a fixed range, and a certain degree of information loss is caused. The one-dimensional self-adaptive convolution kernel convolution layer can perform feature extraction on a plurality of receptive fields through parallel convolution of a plurality of convolution kernels with different sizes, and finally performs weighted summation, so that more and more comprehensive nonlinear features are obtained, and the accuracy of the output ABP waveform is improved.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is an overall frame diagram of a convolutional neural network of the present invention;
FIG. 3 is a block diagram of a downsampling module of the present invention;
FIG. 4 is a block diagram of a one-dimensional adaptive convolution kernel convolution layer of the method of the present invention;
fig. 5 is a block diagram of an upsampling module according to the present invention.
Detailed Description
In this example, an arterial blood pressure signal acquisition method based on pulse wave photoplethysmography, the specific flow is shown in fig. 1, and includes the following steps:
step one, constructing a neural network taking Wave-U-Net as a basic framework, wherein the neural network comprises a downsampling path, a middle layer, an upsampling path and an output layer, and the overall structure of the network is shown in figure 2;
step 1.1, the downsampling path is formed by cascading M downsampling modules, which are respectively marked as Down Block 1 ,...,DownBlock m ,...,DownBlock M The method comprises the steps of carrying out a first treatment on the surface of the Wherein, down Block m Represents an M-th level downsampling module, m=1, 2,; down block of m-th level downsampling module m DSKConv1d as m-th level one-dimensional downsampling adaptive convolution kernel convolution layer m And Downsampling layer Downsampling at level m m The composition, m=8 in this example, and the structure of the M-th downsampling module is shown in fig. 3;
m-th level one-dimensional self-adaptive convolution kernel convolution layer DSKConv1d m Comprising the following steps: an mth separate convolution layer, an mth compressed re-excitation processing layer and an mth weighted combination layer, and an mth one-dimensional downsampling adaptive volumeDeconvolution layer DSKConv1d m The output of (a) is denoted as DSKreult m ；
M-th level one-dimensional adaptive convolution kernel convolution layer DSKConv1d m As shown in fig. 4, the mth separate convolution layer includes: x parallel separate convolutions, X batch normalization layers and X LeakyReLU activation function layers with a slope alpha; wherein the size of the convolution kernel of the xth separate convolution is F x And a convolution kernel F which is separately convolved with the (x+1) th one x+1 The difference between is k, x=0, 1,..x-1; let initialize F 0 =p; in this example x=2, α=0.1, p=15, r 1 ＝…＝r m ＝…＝r 8 =2, convolution step size 1, adaptive padding zero;
when m=1, the first-stage one-dimensional downsampling adaptive convolution kernel convolution layer DSKConv1d in the downsampling path 1 The number of input channels is 3 and the number of output channels is 24, when m=2, 3, 8, the m-th level one-dimensional downsampling is performed on the adaptive convolutional layer DSKConv1d m The number of input channels is 24× (m-1), the number of output channels is 24×m, m=2, 3,..8;
DSKConv1d with m-th level one-dimensional downsampling adaptive convolution kernel convolution layer m For example, the treatment process is as follows: obtaining an input signal Down result m-1 Then, firstly, a separation convolution process is carried out on the downlink result m-1 Respectively performing two one-dimensional convolutions, batch normalization and LeakyReLU, wherein convolution kernels are respectively 15 and 20, and obtaining two characteristic graphs U m1 And U m2 The method comprises the steps of carrying out a first treatment on the surface of the U is set to m1 And U m2 Adding elements to obtain U m For U m Averaging in the channel dimension to obtain statistical information S of each channel m The method comprises the steps of carrying out a first treatment on the surface of the Then the compression re-excitation process is carried out, S m Compressing the channel number to half of the original channel number through a full connection layer, a batch normalization layer and a LeakyReLU activation function, and completing dimension compression to obtain a feature vector Z m ，Z m The channel number is reduced through two full connection layers respectively to finish excitation, and then the two vectors are subjected to softmax operation to obtain two attention vectors A m And B m The method comprises the steps of carrying out a first treatment on the surface of the Finally, DSKConv1d m Output of DSKreult m ＝A m ·U m1 +B m ·U m2 ；
Step 1.2, the up-sampling path is formed by cascading M up-sampling modules, which are respectively marked as UpBlock 1 ,...,UpBlock m ,...,UpBlock M The specific structure of the m-th up-sampling module is shown in fig. 5; wherein UpBlock m Representing an mth level upsampling module; m-th level up-sampling module UpBlock m By Upsampling layer Upsampling at the mth level m And m-th level one-dimensional up-sampling adaptive convolution kernel convolution layer USKConv1d m Composition; and m-th level one-dimensional up-sampling adaptive convolution kernel convolution layer USKConv1d m And m-th level one-dimensional downsampling adaptive convolution kernel convolution layer DSKConv1d in downsampling path m Similarly, m-th level one-dimensional upsampling adaptive convolution kernel convolution layer USKConv1d m The output of (2) is denoted as USKreult m ；
M-th level Upsampling layer Upsampling m The time resolution of the signal is changed into the original r times by adopting a linear interpolation method, and the up-sampling layer of the mth level is Upsampling m The output of (2) is denoted as Upresult m ；
M-th level one-dimensional upsampling adaptive convolution layer USKConv1d in upsampling path m Is structured with m-th level one-dimensional downsampling adaptive convolution layer DSKConv1d m Is completely consistent, except that the m-th level one-dimensional up-sampling self-adaptive convolution layer USKConv1d m Each convolution of the separate convolution layers has an input channel number of 48×m and an output channel number of 24×m, m=1, 2,..8;
step 1.3, the middle layer is a one-dimensional convolution layer, which comprises the following steps: a one-dimensional convolution with a convolution kernel size q, a batch normalization layer, and a LeakyReLU activation function layer with a slope alpha; the output of the intermediate layer is denoted as Mresult;
the output layer is another one-dimensional convolution layer, comprising: a one-dimensional convolution with a convolution kernel of size q, a batch normalization layer, and a tanh activation function layer; the output of the output layer is denoted as ABP est Namely arterial blood pressure signals;
in this example, q=15, the number of input channels and the number of output channels of the middle layer are 192, the number of input channels of the output layer is 27, and the number of output channels is 1;
step 1.4, inputting the input signals into a neural network, sequentially processing the down sampling path, the middle layer, the up sampling path and the output layer, and outputting arterial blood pressure signals ABP est The method comprises the steps of carrying out a first treatment on the surface of the Wherein:
when m=1, 2,..m-1, the M-th stage downsampling module in the downsampling path downsamples down block m In (1) an m-th level one-dimensional downsampling adaptive convolution kernel convolution layer DSKConv1d m Output of DSKreult m Except Downsampling as an mth level Downsampling layer m In addition to the input of (2), the up-sampling module UpBlock of the (m+1) th stage m+1 In (1) m+1st stage one-dimensional upsampling adaptive convolution kernel convolution layer USKConv1d m Output USKreult of (C) m+1 Together as an m-th up-sampling module UpBlock m Middle m-th level one-dimensional self-adaptive convolution kernel convolution layer USKConv1d m Is input to the computer;
m-th downsampling module Down Block in downsampling path M Middle M-th level one-dimensional downsampling self-adaptive convolution kernel convolution layer DSKConv1d M Output of DSKreult M Except Downsampling as an M-th Downsampling layer M In addition to the input of (2), the M-stage up-sampling module UpBlock is used together with the output Mresult of the middle layer M Mid M-th level one-dimensional up-sampling self-adaptive convolution kernel convolution layer USKConv1d M Is input to the computer;
step 1.5, establishing a loss function L by using the formula (1) and taking the loss function L as an optimization target of the convolutional neural network:
L＝L mse +λL max (1)
in the formula (1), L mse The mean square loss is obtained by the formula (2); l (L) max A maximum loss function obtained by the formula (3); lambda is L max Weights of (2); l (L) mse The output of the network can be made as close as possible to the reference signal as a whole, but this is not enough to meet our requirements for ABP waveform details, so we add L that penalizes the details of the waveform max To ensure whether macroscopicThe trend is also a local feature, and the generated waveform of the network is as close as possible to the reference signal. In this example, λ=0.005, so that the two terms can be unified in dimension and represented by L mse L is the main penalty term max Is a correction term to achieve an optimal effect;
in the formula (2), l is the number of signal sample points;
ABP waveform ABP representing network generation est Is the first sample point of (2); />
Representing the first sample point of the reference ABP waveform;
in the formula (3), max is a maximum value taking function;
training a network to obtain an optimal model for converting pulse wave signals into arterial blood pressure signals;
step 2.1, acquiring a fingertip pulse wave signal P with a processing window length L and an arterial blood pressure signal A acquired by an invasive blood pressure sensor from a database;
in this example, l=1024. The raw data used is from the subject's data in the public dataset MIMIC II including both pulse wave signals, arterial blood pressure signals. The acquisition scene of the MIMINIC II data set is an intensive care unit, the pulse wave signal acquisition mode is contact type finger pulse measurement, and the arterial blood pressure signal acquisition mode is invasive measurement. The sampling rate of all signals is 125Hz.
Step 2.2, preprocessing data;
step 2.2.1, performing trend removal and normalization processing on the pulse wave signal P to obtainThe preprocessed pulse wave signal
Then +.>
Obtaining a first derivative velocity plethysmograph signal by taking the first and second order differences>
And second derivative acceleration plethysmograph wave +.>
Finally, the preprocessed pulse wave signal is +.>
First derivative velocity plethysmographic signal +.>
And second derivative acceleration plethysmograph wave +.>
Merging the signals into a 3 xL matrix and taking the matrix as an input signal IN of a neural network;
in the present embodiment, the trend of the pulse wave signal is removed by using (4) to obtain a trend-removed pulse wave signal P d The trending mode can keep the time domain and frequency domain characteristics of the pulse wave signals and remove the low-frequency noise signals in the signals; in the process of collecting pulse wave signals, tiny periodic movements such as respiration of a person to be collected can interfere with frequency components of the pulse waves, so that low-frequency components in the pulse wave signals are removed by using trend removal; normalizing the pulse wave signal by using the method (6) to obtain a preprocessed pulse wave signal
The normalization method can retain pulse wave signalsThe time domain and frequency domain characteristics of the signal are converted into a mean value of 0, the standard deviation of 1 and the amplitude of the signal is within the interval of [0,1]The normalized signals are convenient for training the neural network;
in the formula (4), I is 1024-order unit arrays, the constant lambda is set to 10, D 2 ∈R (x-3)×(x-1) A second-order differential matrix shown in the formula (5);
in the formula (6), mu is the pulse wave signal P after trend removal d Sigma is the mean value of the pulse wave signal P after trend removal d Standard deviation of (2);
step 2.2.2, normalizing the arterial blood pressure signal A to obtain ABP ref And ABP is carried out ref A synchronization reference signal of length L as an input signal IN; in this example, the amplitude of the arterial blood pressure signal is compressed to 1/200 of the original; the activation function of the output layer in the network is the tanh function, so that the amplitude of the network output signal can only be in the interval 0,1]Between them; the amplitude is directly scaled so as to facilitate the training of the network, and the result can be restored very conveniently in actual use;
step 2.3, training the network;
input signal IN is sent into neural network, and corresponding reference signal ABP ref Calculating a loss function L as an optimization target, enabling the loss function L to tend to be stable in the training process, and stopping training and obtaining an optimal model when the absolute value of the variation delta L of the loss function L is always smaller than a set threshold value theta;
in this example, θ=4×10 -5 The method comprises the steps of carrying out a first treatment on the surface of the Use of optimaWhen the model is tested, only a matrix with the dimension of 3 multiplied by 1024, which is formed by combining pulse wave signals, VPG signals and APG signals, is used as input, so that corresponding ABP waveforms can be obtained, and the obtained ABP waveforms are used for extracting cardiovascular related parameters and diagnosing cardiovascular diseases;
step 2.4, arterial blood pressure signal post-processing:
arterial blood pressure signal ABP output by optimal model est The original dimension is restored to obtain the converted arterial blood pressure signal, and in the example, the dimension can be restored by directly multiplying the amplitude of the output waveform of the network by 200.
Claims (1)
1. An arterial blood pressure signal acquisition method based on pulse wave photoplethysmography is characterized by comprising the following steps:
step one, constructing a neural network taking Wave-U-Net as an infrastructure, which comprises the following steps: a downsampling path, an intermediate layer, an upsampling path, and an output layer;
step 1.1, the downsampling path is formed by cascading M downsampling modules, and the M downsampling modules are respectively marked as Down Block 1 ,...,DownBlock m ,...,DownBlock M The method comprises the steps of carrying out a first treatment on the surface of the Wherein, down Block m Represents an M-th level downsampling module, m=1, 2,; down block of m-th level downsampling module m DSKConv1d as m-th level one-dimensional downsampling adaptive convolution kernel convolution layer m And Downsampling layer Downsampling at level m m Composition;
m-th level one-dimensional self-adaptive convolution kernel convolution layer DSKConv1d m Comprising the following steps: an mth separate convolution layer, an mth compressed re-excitation processing layer and an mth weighted combination layer, and an mth one-dimensional downsampling adaptive convolution kernel convolution layer DSKConv1d m The output of (a) is denoted as DSKreult m ；
The mth separate convolution layer includes: x parallel separate convolutions, X batch normalization layers and X LeakyReLU activation function layers with a slope alpha; wherein the size of the convolution kernel of the xth separate convolution is F x And a convolution kernel F which is separately convolved with the (x+1) th one x+1 The difference between is k, x=0, 1,..x-1;let initialize F 0 ＝p；
Downsampling layer Downsampling of m-th level m For changing the time resolution of the signal to be original
r represents a scaling factor; downsampling layer Downsampling of m-th level m The output of (2) is denoted as Down result m ；
Step 1.2, the up-sampling path is formed by cascading M up-sampling modules, and the M up-sampling modules are respectively marked as UpBlock 1 ,...,UpBlock m ,...,UpBlock M The method comprises the steps of carrying out a first treatment on the surface of the Wherein UpBlock m Representing an mth level upsampling module; m-th level up-sampling module UpBlock m From Upsampling at the mth level m Sampling layer and m-th level one-dimensional up-sampling self-adaptive convolution kernel convolution layer USKConv1d m Composition; the m-th level one-dimensional up-sampling self-adaptive convolution kernel convolution layer USKConv1d m And m-th level one-dimensional downsampling adaptive convolution kernel convolution layer DSKConv1d in downsampling path m Similarly, m-th level one-dimensional upsampling adaptive convolution kernel convolution layer USKConv1d m The output of (2) is denoted as USKreult m ；
M-th level Upsampling layer Upsampling m The time resolution of the signal is changed into the original r times by adopting a linear interpolation method, and the up-sampling layer of the mth level is Upsampling m The output of (2) is denoted as Upresult m ；
Step 1.3, the intermediate layer is a one-dimensional convolution layer, comprising: a one-dimensional convolution with a convolution kernel size q, a batch normalization layer, and a LeakyReLU activation function layer with a slope alpha; the output of the intermediate layer is denoted as Mresult;
the output layer is another one-dimensional convolution layer, comprising: a one-dimensional convolution with a convolution kernel of size q, a batch normalization layer, and a tanh activation function layer; the output of the output layer is marked as ABP est Namely arterial blood pressure signals;
step 1.4, inputting an input signal into the neural network and sequentially processing the down-sampling path, the middle layer, the up-sampling path and the output layerAfter that, arterial blood pressure signal ABP is output est The method comprises the steps of carrying out a first treatment on the surface of the Wherein:
when m=1, 2,..m-1, the M-th stage downsampling module in the downsampling path downsamples down block m In (1) an m-th level one-dimensional downsampling adaptive convolution kernel convolution layer DSKConv1d m Output of DSKreult m Except Downsampling as an mth level Downsampling layer m In addition to the input of (2), the up-sampling module UpBlock of the (m+1) th stage m+1 In (1) m+1st stage one-dimensional upsampling adaptive convolution kernel convolution layer USKConv1d m Output USKreult of (C) m+1 Together as an m-th up-sampling module UpBlock m Middle m-th level one-dimensional self-adaptive convolution kernel convolution layer USKConv1d m Is input to the computer;
m-th downsampling module Down Block in downsampling path M Middle M-th level one-dimensional downsampling self-adaptive convolution kernel convolution layer DSKConv1d M Output of DSKreult M Except Downsampling as an M-th Downsampling layer M In addition to the input of (2), the M-stage up-sampling module UpBlock is used together with the output Mresult of the middle layer M Mid M-th level one-dimensional up-sampling self-adaptive convolution kernel convolution layer USKConv1d M Is input to the computer;
step 1.5, establishing a loss function L by using the formula (1) and taking the loss function L as an optimization target of the convolutional neural network:
L＝L mse +λL max (1)
in the formula (1), L mse Is the mean square loss and is obtained by the formula (2); l (L) max Is a maximum loss function and is obtained by the formula (3); lambda is L max Weights of (2);
in the formula (2), L is the number of signal sample points;
arterial blood pressure waveform ABP representing network generation est Is the first sample point of (2); />
A first sample point representing a reference arterial blood pressure waveform;
in the formula (3), max is a maximum value taking function;
training a network to obtain an optimal model for converting pulse wave signals into arterial blood pressure signals;
step 2.1, acquiring a fingertip pulse wave signal P with a processing window length L and an arterial blood pressure signal A acquired by an invasive blood pressure sensor from a database;
step 2.2, preprocessing data;
step 2.2.1, performing trend removal and normalization treatment on the pulse wave signal P to obtain a preprocessed pulse wave signal
Then +.>
Obtaining a first derivative velocity plethysmograph signal by taking the first and second order differences>
And second derivative acceleration plethysmograph wave +.>
Finally, the preprocessed pulse wave signal is +.>
First derivative velocity plethysmographic signal +.>
And second derivativeAcceleration plethysmography wave->
Merging the signals into a 3 xL matrix and taking the matrix as an input signal IN of a neural network;
step 2.2.2, normalizing the arterial blood pressure signal A to obtain a normalized arterial blood pressure signal ABP ref And normalized arterial blood pressure signal ABP ref A synchronization reference signal of length L as an input signal IN;
step 2.3, training the network;
input signal IN is sent into neural network, and corresponding reference signal ABP ref Calculating a loss function L as an optimization target, enabling the loss function L to tend to be stable in the training process, and stopping training and obtaining an optimal model when the absolute value of the variation delta L of the loss function L is always smaller than a set threshold value theta;
step 2.4, arterial blood pressure signal post-processing:
arterial blood pressure signal ABP output by optimal model est And the amplitude of the arterial blood pressure signal is restored to the original dimension, so that the arterial blood pressure signal after conversion is obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110184194.5A CN112914527B (en) | 2021-02-08 | 2021-02-08 | Arterial blood pressure signal acquisition method based on pulse wave photoplethysmography |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110184194.5A CN112914527B (en) | 2021-02-08 | 2021-02-08 | Arterial blood pressure signal acquisition method based on pulse wave photoplethysmography |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112914527A CN112914527A (en) | 2021-06-08 |
| CN112914527B true CN112914527B (en) | 2023-06-30 |
Family
ID=76169702
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110184194.5A Active CN112914527B (en) | 2021-02-08 | 2021-02-08 | Arterial blood pressure signal acquisition method based on pulse wave photoplethysmography |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112914527B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113693575A (en) * | 2021-08-06 | 2021-11-26 | 武汉理工大学 | PPG blood pressure detection method based on lightweight neural network |
| CN113674376B (en) * | 2021-08-23 | 2024-03-05 | 合肥工业大学 | Ultrasonic plane wave composite imaging method based on deep learning |
| CN114224303A (en) * | 2021-10-19 | 2022-03-25 | 安徽大学 | Noninvasive blood pressure prediction method and system |
| CN114145725B (en) * | 2022-02-08 | 2022-05-06 | 广东工业大学 | PPG sampling rate estimation method based on noninvasive continuous blood pressure measurement |
| CN114818910B (en) * | 2022-04-22 | 2023-04-28 | 北京邮电大学 | Non-contact blood pressure detection model training method, blood pressure detection method and device |
| CN114788691B (en) * | 2022-06-20 | 2022-10-04 | 广东工业大学 | Blood pressure prediction method and system based on multi-channel gram angular field |
| CN116383617B (en) * | 2023-04-21 | 2023-09-22 | 复旦大学 | Intelligent blood pressure detection method and system based on pulse wave waveform characteristics |
| CN116400264B (en) * | 2023-06-09 | 2023-08-18 | 广东工业大学 | Inverter open-circuit fault diagnosis method and system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108498089A (en) * | 2018-05-08 | 2018-09-07 | 北京邮电大学 | A kind of noninvasive continuous BP measurement method based on deep neural network |
| WO2020007041A1 (en) * | 2018-07-03 | 2020-01-09 | 京东方科技集团股份有限公司 | Blood pressure measurement apparatus and method, electronic device, and computer readable storage medium |
| CN110840428A (en) * | 2019-11-29 | 2020-02-28 | 苏州大学 | Noninvasive blood pressure estimation method based on one-dimensional U-Net network |
| CN111493850A (en) * | 2020-04-13 | 2020-08-07 | 中国科学院深圳先进技术研究院 | Blood pressure measuring method and device based on deep neural network |
-
2021
- 2021-02-08 CN CN202110184194.5A patent/CN112914527B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108498089A (en) * | 2018-05-08 | 2018-09-07 | 北京邮电大学 | A kind of noninvasive continuous BP measurement method based on deep neural network |
| WO2020007041A1 (en) * | 2018-07-03 | 2020-01-09 | 京东方科技集团股份有限公司 | Blood pressure measurement apparatus and method, electronic device, and computer readable storage medium |
| CN110840428A (en) * | 2019-11-29 | 2020-02-28 | 苏州大学 | Noninvasive blood pressure estimation method based on one-dimensional U-Net network |
| CN111493850A (en) * | 2020-04-13 | 2020-08-07 | 中国科学院深圳先进技术研究院 | Blood pressure measuring method and device based on deep neural network |
Non-Patent Citations (1)
| Title |
|---|
| 基于脉搏波和小波神经网络的血压预测;刘艳萍;李杰;金菲;崔彤;;河北工业大学学报(第03期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112914527A (en) | 2021-06-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112914527B (en) | Arterial blood pressure signal acquisition method based on pulse wave photoplethysmography | |
| CN112587153B (en) | End-to-end non-contact atrial fibrillation automatic detection system and method based on vPPG signal | |
| WO2020228420A1 (en) | Denoising autoencoder training method, electrocardiography signal denoising method, and apparatuses | |
| WO2021164349A1 (en) | Blood pressure prediction method and apparatus based on photoplethysmography signal | |
| WO2023185873A1 (en) | Non-cuff blood pressure measuring apparatus based on multi-stage multi-modal learning and method | |
| CN113116321A (en) | Non-invasive continuous blood pressure measuring system based on PSO-GRNN neural network | |
| CN114628020A (en) | Remote plethysmography signal detection model construction and detection method, device and application | |
| CN114569096A (en) | Non-contact continuous blood pressure measuring method and system based on video stream | |
| CN114052693A (en) | Heart rate analysis method, device and equipment | |
| Sivanjaneyulu et al. | Cnn based ppg signal quality assessment using raw ppg signal for energy-efficient ppg analysis devices in internet of medical things | |
| CN114145725B (en) | PPG sampling rate estimation method based on noninvasive continuous blood pressure measurement | |
| CN114668375A (en) | Cuff-free blood pressure prediction system based on deep neural network model | |
| CN116369877A (en) | Noninvasive blood pressure estimation method based on photoelectric volume pulse wave | |
| CN114947782A (en) | System and method for central arterial pressure waveform reconstruction based on PPG signal | |
| Kuang et al. | Remote photoplethysmography signals enhancement based on generative adversarial networks | |
| Sivanjaneyulu et al. | Compressive Sensing-Based Automatic PPG Signal Quality Assessment Using CNN for Energy-Constrained Medical Devices | |
| CN115905819B (en) | rPPG signal generation method and device based on generation countermeasure network | |
| Anbukarasu et al. | Tiny-HR: Towards an interpretable machine learning pipeline for heart rate estimation on edge devices | |
| CN114947792B (en) | Video-based physiological signal measurement and enhancement method | |
| CN114788691B (en) | Blood pressure prediction method and system based on multi-channel gram angular field | |
| CN116889388B (en) | Intelligent detection system and method based on rPPG technology | |
| CN116035548B (en) | Method and device for detecting heart rhythm state, intelligent wearable device and storage medium | |
| CN117898691A (en) | Non-contact heart rate detection method based on KDN-WTRFA | |
| US20230148880A1 (en) | Method and system for determining cardiovascular parameters | |
| Moghaddam Mansouri et al. | Systolic Blood Pressure Estimation from Electrocardiogram and Photoplethysmogram Signals Using Convoluional Neural Networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |